The $19B "Nuclear AI" Energy Startup That Couldn't Sign a Single Client (financialpost.com) 44
"Nuclear AI startup" Fermi had hoped to build power plants generating 17 gigawatts of electricity, remembers Bloomberg, "three times the amount typically consumed by New York City."
Hyperscalers could install their data centers on the site itself and tap directly into that power, which would come first from natural gas turbines and later from nuclear reactors. The pitch ticked so many boxes — artificial intelligence, nuclear energy, political connections — that some investors found it irresistible. Fermi went public in October worth more than $19 billion in market value, despite reporting no revenue or signed customers.
Now, the startup's board has fired its top executive, Toby Neugebauer, after months of negotiations failed to secure a single client. Chief Financial Officer Miles Everson left as well... Fermi's stock, meanwhile, has tumbled 84% from its peak. The company's more than 5,000-acre site in the Texas panhandle — dubbed Project Matador, or the President Donald J. Trump Advanced Energy and Intelligence Campus — remains mostly unfinished. And some analysts see a cautionary tale of the market's AI enthusiasm running ahead of reality, with investors betting on companies whose grand projects may never get built...
The idea of giving data centers their own, dedicated power supply not dependent on the grid may sound tempting, but former US Department of Energy official Jigar Shah said banks don't want to finance it. The grid, drawing power from many sources, is more reliable than a handful of expensive, on-site plants, he said. He considers Fermi a failure "of monumental proportions" and says similar, off-grid data center projects elsewhere deserve more skepticism than they've received... "We're allowing these types of projects to continue to be viewed as viable when they most certainly are not," said Shah, who ran the department's Loan Programs Office during the Biden administration....
"It was a piece of dirt with a dream," an investor who visited the site in February told the short sellers, Fuzzy Panda Research.
Now, the startup's board has fired its top executive, Toby Neugebauer, after months of negotiations failed to secure a single client. Chief Financial Officer Miles Everson left as well... Fermi's stock, meanwhile, has tumbled 84% from its peak. The company's more than 5,000-acre site in the Texas panhandle — dubbed Project Matador, or the President Donald J. Trump Advanced Energy and Intelligence Campus — remains mostly unfinished. And some analysts see a cautionary tale of the market's AI enthusiasm running ahead of reality, with investors betting on companies whose grand projects may never get built...
The idea of giving data centers their own, dedicated power supply not dependent on the grid may sound tempting, but former US Department of Energy official Jigar Shah said banks don't want to finance it. The grid, drawing power from many sources, is more reliable than a handful of expensive, on-site plants, he said. He considers Fermi a failure "of monumental proportions" and says similar, off-grid data center projects elsewhere deserve more skepticism than they've received... "We're allowing these types of projects to continue to be viewed as viable when they most certainly are not," said Shah, who ran the department's Loan Programs Office during the Biden administration....
"It was a piece of dirt with a dream," an investor who visited the site in February told the short sellers, Fuzzy Panda Research.
"A startup in hot industry- (Score:2)
Re: (Score:1, Offtopic)
Honestly Solyndra was blown out of proportion (Score:4, Interesting)
It's almost as if there was a coordinated attempt to discredit solar energy by powerful people who had nearly limitless money and a vested interest in making sure the public had a bad taste in their mouth regarding solar power...
It's a good thing crazy conspiracy theories like that don't happen in the real world. Now if you excuse me I'm going to go play some Nintendo games. I'm thinking I'm going to play some contra.
Re:Honestly Solyndra was blown out of proportion (Score:5, Interesting)
The government lost about half a billion. Not chump change but a rounding error if you're talking a serious attempt to get off oil.
Solyndra was just one part of the program. They made money overall.
Thanks I always forget about that (Score:4, Interesting)
It wasn't really until this last couple of years that I really realized just how thorough and complete the media capture is.
Re:Thanks I always forget about that (Score:4, Insightful)
Oh, absolutely. It was Republicans, specifically with the backing of the Koch brothers and their oil interests that tried to blow it up into a big scandal instead of just an idea that didn't work out. Mitt Romney tried to use it as a campaign issue as well. Ironically, one of their big things was about how the government should not be picking winners and losers in the market, should not be investing in private business, etc. Now of course we have the trump administration taking stakes in corporations and creating a sovereign wealth fund and the same people who railed against the program that invested in Solyndra are cheering it on.
Solyndra story was always a lie. (Score:5, Insightful)
Sigh. Yes. Yes, I do, because for TWELVE FREAKING YEARS, I've been having to find and copy this excerpt from the NYT story on it, November 16, 2014:
"Here’s another: Remember Solyndra? It was a renewable-energy firm that borrowed money using Department of Energy guarantees, then went bust, costing the Treasury $528 million. And conservatives have pounded on that loss relentlessly, turning it into a symbol of what they claim is rampant crony capitalism and a huge waste of taxpayer money.
Defenders of the energy program tried in vain to point out that anyone who makes a lot of investments, whether it’s the government or a private venture capitalist, is going to see some of those investments go bad. For example, Warren Buffett is an investing legend, with good reason — but even he has had his share of lemons, like the $873 million loss he announced earlier this year on his investment in a Texas energy company. Yes, that’s half again as big as the federal loss on Solyndra.
The question is not whether the Department of Energy has made some bad loans — if it hasn’t, it’s not taking enough risks. It’s whether it has a pattern of bad loans. And the answer, it turns out, is no. Last week the department revealed that the program that included Solyndra is, in fact, on track to return profits of $5 billion or more."
Re: (Score:2)
Re: (Score:1)
Wait'll you hear about the current administration.
https://www.npr.org/2025/08/03... [npr.org]
Re: (Score:3)
It's not even part of the argument, it's a total non-argument. You're saying "its true because it's true" which is just mind projection. You'd have to show this level of investment had a higher failure rate (I think for the DOE program it was 3%?) than the private market. Does it?
Re: (Score:2)
Remember Solyndra?
Yup and he was a badass -- oh wait ... sorry, that was Solen'ya [fandom.com]. :-)
Re:"A startup in hot industry- (Score:5, Interesting)
Remember Solyndra?
Solyndra was a real product though. It came out just as solar panels were getting cheap and couldn't compete with flat panels. They did do a number of things to bolster their apparent success, but they clearly were not created just to ride a financial bubble and bilk investors. Looking at their product, you can see how they could have reasonably thought they had a shot in the market.
Looking at Fermi, it is hard to see how anyone could have reasonably thought they had a shot. They never had a tangible product. On the other hand, they did make some effort to set up generation.
In the end, Solyndra sold actual product and Fermi didn't. Maybe they can both fall under the category of being aspirational but unsuccessful, but the sheer lack of realism in the Fermi model makes me rate it much, much higher on the scam scale than I would Solyndra.
Re: (Score:3)
I remember it was part of the same gov't program that funded Tesla.
I also remember that only about 2% of the funded programs at the time went bankrupt
So the scam was incredibly obvious? (Score:2, Redundant)
Even more obvious than what the nuclear bros usually push, I take it. Well. With the demented things investors are willing to finance these days, they must have really, really, really screwed up their pitch.
Re: (Score:2)
the President Donald J. Trump Advanced Energy and Intelligence Campus
how could it have been anything other than a grift?
Re: (Score:2)
Indeed. But the surprising thing is that the investors apparently all saw and understood that. The mountain of evidence must have gotten impossible to ignore.
Re: (Score:2)
Given that it was
the President Donald J. Trump Advanced Energy and Intelligence Campus
how could it have been anything other than a grift?
Oh, ye of little faith. Just you watch...the company will have a dozen customers in two weeks.
It's one thing to run a scam (Score:2)
I mean up until about halfway through Trump's third term. But then anything goes.
'cause they all know it's a fad (Score:2)
Nuclear power stations take decades to build. The customer base knows this whole AI thing isn't going to really pan out even as they sell it. They need to make their money as soon as possible before the curtain falls, and that will be much sooner than this power solution will be ready.
Named to fail (Score:1)
Damn, talk about self-sabotage. Maybe if they stopped telling all their potential clients that they are a scam, they might get a sale.
Lets set a few things straight (Score:5, Interesting)
Their plan was to buy real estate to be dedicated for data centers, and paired with that they would provide "behind-the-meter" power; in essence they would not only lease the shell buildings as data centers but also sell power generated. They were talking about nuclear, but I don't think they were that stupid. A modern stick built nuclear reactor can take 5-8 years after a multi-year permitting process, so this just wasn't going to happen in any reasonable time. They were also planning on putting on-site natural gas and solar, which are vastly quicker to set up. So part of the distribution to investors was also in the form of energy sales direct to the client AI companies taking over.
The CEO was fired for cause per the Board's own statements, but the big issue is he had no clients and now 6 months after going public they still had no clients, so even if they started building they had no one to act as a customer.
So in the end, it's a simple act of a guy with good pitch to investors but bad at execution, with an idea that maybe on paper was good (leasing land with dedicated power to data center places), but came down to bad math. He was never going to be able to provide the 11GW he proposed without an insane amount of financing (estimates $120B was needed to build that out), he lost a deal with Amazon when he couldn't' give them confidence the power would be there, and he couldn't get favorable terms to make it happen. he should have bent over backwards to get Amazon in there as an anchor client to prove the model and go from there.
Re: (Score:2)
I'm hugely surprised at this, really. NO clients at all? Sigh ...
Because investors don't get advice from Fox (Score:5, Interesting)
Investors can watch the nuclear industry trying hard with SMRs and thorium and traveling wave and pebble-bed, and NOT see industry getting the construction cost down below $10/watt. One hears tell that China is kicking reactors out the door in sixty months flat, at $4.50/watt. But then, China is an unreliable narrator about the greatness of China, and may be doing awful things to local environments. Even if they lean on our governments to ease up on nuclear (Trump sure would), halving the cost to $5/watt seems - well, unlikely: RISKY. And capital is a coward.
Meanwhile, investors are watching batteries and renewables get cheaper, steadily, are monitoring announcements from pilot plants and upstream to the labs, and must feel pretty confident that prices will KEEP getting cheaper, significantly so in the five-year minimum construction time for a reactor.
Nuclear is great, I was a champion for 50 years. But I can read an accounting ledger, and nuclear has been beaten.
Re: (Score:2)
It doesn't work that way. You need baseline power, pure and simple. It's either coal/oil/gas, or it's nuclear. Nothing else works, and nuclear is the best out of these options.
Re: (Score:2)
First, it is called "baseload". And second, nuclear is so bad at baseload that 70% is the absolute maximum a grid can tolerate before it becomes unstable. And you need fast regulation energy in the size of your largest nuke, which is very expensive. Because while wind and solar can be planned, a nuke SCRAMing comes with no warning at all.
Re: (Score:2)
What happens when someone hits the EPO at a multi-gigawatt datacenter?
AI Data Centers are hedging their risky bet (Score:2)
Re: (Score:2)
It is the other way round. The whole base load concept only exists because coal and nuclear suck at load following. If all power generation is capable of load following, base load is moot.
Re:Because investors don't get advice from Fox (Score:4)
The term is "dispatch". A dispatchable power source is able to do load following - batteries, hydro, and natural gas plants are dispatchable in they can react to changing load conditions in minutes. Minutes is fine because of another concept called "inertia" - high inertia power sources resist changes to loads much better than low inertia sources. This is why the grid only needs minute-level responses - things like hydro plants have the power of flowing water providing a ton of inertia - a sudden jump in generator load is met with a sudden increase in force on the generator turbine (see water hammer) which keeps the turbine spinning at a constant rate.
Non-dispatchable sources are ones like coal, nuclear, solar and wind. This is because their output cannot adjust tot he load demanded on demand. Some like nuclear and coal, take hours to adjust. Others like solar and wind simply cannot be adjusted - you get what you get.
Now, we can compensate for non-dispatchable sources - nuclear and coal we run at a level that guarantees they will not power the entire grid should demand drop - we run them under what the grid will demand, then top it up with dispatchable sources.
Solar and wind, we take the opposite approach - we can't ask them to produce more, but we can ask them to produce less, so output of solar and wind is curtailed to meet demand.
It's a whole control system where you have base load supplied by nuclear, and then wind and solar producing as much as they can as it's the cheapest. You then ride out the shortfalls with batteries, hydro or natural gas where you can curtail the output quickly.
And yes, you better hope no one hits the EPO because that would force an emergency shutdown for those nuclear reactors but even then that's still half a day's worth of power that needs to go somewhere (even a SCRAM still needs to dissipate the heat and that takes 6-12 hours to do even in emergency shutdown) Or you have to phase it down - if a fire happens you turn off the equipment in that room and isolate the rest of it from the room so the load doesn't shift too badly. At the same time you reduce the power output of the reactors in anticipation of needing to shut down more reactors and let the residual heat power the rest of the data center until there's not enough then spread the blackout outwards from the affected area.
You might rely on the grid a bit but it's likely to be a bottleneck in that the grid tie can't handle the full load of the datacenter which also means it can't handle the full output of the reactors but it can absorb some of it.
Re: (Score:2)
It's funny how concepts like that get so anchored in the gestalt. Mental shortcuts that short circuit thought.
Re: (Score:2)
It surprising that people still get this completely wrong even though it is not terribly difficult to understand. Base load is the level of power you always need even when demand is low, and base plants are plants that can provide this base load of power cheaply when they are always running with constant output. So in the past this could be plants that are very expensive to build but cheap to operate such as nuclear. But once you have more variation on the grid from sources that are also very cheap such as
Re: (Score:2)
It's funny how when a Chinese company makes a really good car, it's just a knock-off using stolen technology and the fair price is heavily subsidized by the CCP. When a Chinese company names a nuclear reactor, it's because they have very sensible nuclear regulations and have developed low cost manufacturing techniques, without any free government money so of course we should steal, I mean copy it.
The reality is that the government there got behind nuclear to give the industry certainty. The technology was l
No shit (Score:2)
Investors are not smart people and do not make smart decisions.
If Trump's investing in it... (Score:1)
If Trump's investing in it you can be all but guaranteed it's going to flop.
color me shocked (Score:4, Funny)
Wow. That's so ... um ... surprising.
Re: (Score:2)
Ah. That is why no investor is willing to support this venture. Makes sense given his business history.
Re: (Score:3)
You have it skewed. The investor crowd loved it and as a result they have lots of cash on hand. In that sense it isn't a scam at all, most of the money is still there and they'll either be more successful under new leadership, or they can wind it down and return the majority of the money to investors.
The story is that there are no customers willing to risk their business on it. The idiot in charge of an investment willing to name their site after President Trump was exactly the sort of person that couldn't
No crypto? No NFT? (Score:2)